Performance Enhancement of Extended AFDX via Bandwidth Reservation - - PowerPoint PPT Presentation

Performance Enhancement of Extended AFDX via Bandwidth Reservation for TSN/BLS Shapers

Ana¨ ıs Finzi, Ahlem Mifdaoui et al. July 3, 2018, RTN’18

1/27

2/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

2/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Up to 500 km of cables

2/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Up to 500 km of cables Heterogeneous network

2/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Up to 500 km of cables Heterogeneous network AFDX

2/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Up to 500 km of cables Heterogeneous network AFDX ARINC 429

2/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Up to 500 km of cables Heterogeneous network AFDX ARINC 429 CAN

2/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Up to 500 km of cables Heterogeneous network AFDX ARINC 429 CAN MIL-STD-1553

3/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Current Avionics Communication Architecture limitations Heterogeneity: high complexity, delays and costs One criticality level: backbone supports only essential traffic Unfair service policy: strong impact of high priorities

3/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Context and Objectives

Current Avionics Communication Architecture limitations Heterogeneity: high complexity, delays and costs One criticality level: backbone supports only essential traffic Unfair service policy: strong impact of high priorities Main Objective

Homogenize avionics communication architecture → Extend the backbone network to support Safety-Critical and Best-EffortTraffics

4/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Avionics Requirements and Challenges

Requirements Predictability : guaranteeing schedulability constraints, i.e. bounded delays respecting deadlines Modularity : minimizing the (re)configuration effort

4/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Avionics Requirements and Challenges

Requirements Predictability : guaranteeing schedulability constraints, i.e. bounded delays respecting deadlines Modularity : minimizing the (re)configuration effort Challenges ցComplexity : Reduce the implementation and configuration effort րFairness : Limit the impact of high priorities on lower ones

5/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Promising Solution

Solutions TTE1 TAS2 PS3 UBS4 BLS5 CBS6 NP-SP7 DRR8 Modularity X X X

• Predictability
• Fairness

X X

• X
• Complexity

X X X X

• Existing solutions vs avionics requirements and challenges

: : X:

1Time Triggered Ethernet 2Time Aware Shaper 3Peristaltic Shaper 4Urgency Based Scheduler 5Burst Limiting Shaper 6Credit-based Shaper 7Non-preemptive Static Priority 8Deficit Round Robin

5/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Promising Solution

Solutions TTE1 TAS2 PS3 UBS4 BLS5 CBS6 NP-SP7 DRR8 Modularity X X X

• Predictability
• Fairness

X X

• X
• Complexity

X X X X

• Existing solutions vs avionics requirements and challenges

: : X: Schedulers

1Time Triggered Ethernet 2Time Aware Shaper 3Peristaltic Shaper 4Urgency Based Scheduler 5Burst Limiting Shaper 6Credit-based Shaper 7Non-preemptive Static Priority 8Deficit Round Robin

5/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Promising Solution

Solutions TTE1 TAS2 PS3 UBS4 BLS5 CBS6 NP-SP7 DRR8 Modularity X X X

• Predictability
• Fairness

X X

• X
• Complexity

X X X X

• Existing solutions vs avionics requirements and challenges

: : X: Schedulers TTTEch

1Time Triggered Ethernet 2Time Aware Shaper 3Peristaltic Shaper 4Urgency Based Scheduler 5Burst Limiting Shaper 6Credit-based Shaper 7Non-preemptive Static Priority 8Deficit Round Robin

5/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Promising Solution

Solutions TTE1 TAS2 PS3 UBS4 BLS5 CBS6 NP-SP7 DRR8 Modularity X X X

• Predictability
• Fairness

X X

• X
• Complexity

X X X X

• Existing solutions vs avionics requirements and challenges

: : X: Schedulers TTTEch IEEE Time Sensitive Networking

1Time Triggered Ethernet 2Time Aware Shaper 3Peristaltic Shaper 4Urgency Based Scheduler 5Burst Limiting Shaper 6Credit-based Shaper 7Non-preemptive Static Priority 8Deficit Round Robin

5/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Promising Solution

Solutions TTE1 TAS2 PS3 UBS4 BLS5 CBS6 NP-SP7 DRR8 Modularity X X X

• Predictability
• Fairness

X X

• X
• Complexity

X X X X

• Existing solutions vs avionics requirements and challenges

: : X:

1Time Triggered Ethernet 2Time Aware Shaper 3Peristaltic Shaper 4Urgency Based Scheduler 5Burst Limiting Shaper 6Credit-based Shaper 7Non-preemptive Static Priority 8Deficit Round Robin

5/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Promising Solution

Solutions TTE1 TAS2 PS3 UBS4 BLS5 CBS6 NP-SP7 DRR8 Modularity X X X

• Predictability
• Fairness

X X

• X
• Complexity

X X X X

• Existing solutions vs avionics requirements and challenges

: : X:

1Time Triggered Ethernet 2Time Aware Shaper 3Peristaltic Shaper 4Urgency Based Scheduler 5Burst Limiting Shaper 6Credit-based Shaper 7Non-preemptive Static Priority 8Deficit Round Robin

5/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Promising Solution

Solutions TTE1 TAS2 PS3 UBS4 BLS5 CBS6 NP-SP7 DRR8 Modularity X X X

• Predictability
• Fairness

X X

• X
• Complexity

X X X X

• Existing solutions vs avionics requirements and challenges

: : X:

→ the Burst Limiting Shaper is the most promising solution

1Time Triggered Ethernet 2Time Aware Shaper 3Peristaltic Shaper 4Urgency Based Scheduler 5Burst Limiting Shaper 6Credit-based Shaper 7Non-preemptive Static Priority 8Deficit Round Robin

6/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Followed Methodology

Specification of an Extended AFDX → Low complexity and few hardware/software modificationsa

a[ERTS2-18] Finzi, A., Mifdaoui et al., ”Mixed-Criticality on the AFDX

Network: Challenges and Potential Solutions”, ERTS 2018

6/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Followed Methodology

Specification of an Extended AFDX → Low complexity and few hardware/software modificationsa

a[ERTS2-18] Finzi, A., Mifdaoui et al., ”Mixed-Criticality on the AFDX

Network: Challenges and Potential Solutions”, ERTS 2018

Formal timing analysis → New Network Calculus model with good tightness a

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed- Criticality Applications: Model and Timing Analysis”, WFCS 2018

6/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Followed Methodology

Specification of an Extended AFDX → Low complexity and few hardware/software modificationsa

a[ERTS2-18] Finzi, A., Mifdaoui et al., ”Mixed-Criticality on the AFDX

Network: Challenges and Potential Solutions”, ERTS 2018

Formal timing analysis → New Network Calculus model with good tightness a

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed- Criticality Applications: Model and Timing Analysis”, WFCS 2018

Performance Enhancement → Bandwidth Reservation methods for TSN/BLS to enhance system schedulability

7/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Outline

1

System Model

2

Bandwidth Reservation Methods

3

Performance Evaluation

4

Conclusion and perspectives

8/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Outline

1

System Model

2

Bandwidth Reservation Methods

3

Performance Evaluation

4

Conclusion and perspectives

9/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Extended AFDX Switch

1-Gigabit AFDX Switch architecture

We consider 3 types of traffics: Safety Critical Traffic (SCT), Rate Constrained (RC), and Best-Effort (BE).

BLS SP SCT RC BE BLS SP SCT RC BE forwarding process Input ports Output ports Configuration table

Proposed switch architecture

9/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Extended AFDX Switch

1-Gigabit AFDX Switch architecture

We consider 3 types of traffics: Safety Critical Traffic (SCT), Rate Constrained (RC), and Best-Effort (BE).

BLS SP SCT RC BE BLS SP SCT RC BE forwarding process Input ports Output ports Configuration table

Proposed switch architecture

10/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Extended AFDX output port

3-classes example: high BLS priority

SP BLS #1 #0 #3 RC class SCT class BE class sets queue priority between {0,2}

11/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Extended AFDX output port

3-classes example: low BLS priority

SP BLS #1 #2 #3 RC class SCT class BE class sets queue priority between {0,2}

12/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Burst Limiting Shaper Parameters

Each BLS credit has 3 parameters: Maximum Level (LM) Resume Level (LR) Reserved Bandwidth (BW) BW is used with the output link capacity C to compute the credit slopes as follows: the sending slope, Isend = (1 − BW ) · C the idle slope, Iidle = BW · C

13/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Burst Limiting Shaper credit evolution

Bursty traffic

t LR LM Transmitted traffic t

13/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Burst Limiting Shaper credit evolution

Bursty traffic

t LR LM Transmitted traffic t

Isend

13/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Burst Limiting Shaper credit evolution

Bursty traffic

t LR LM Transmitted traffic t

Isend

{

non-preemption

13/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Burst Limiting Shaper credit evolution

Bursty traffic

t LR LM Transmitted traffic t

Isend

{

non-preemption

13/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Burst Limiting Shaper credit evolution

Bursty traffic

t LR LM Transmitted traffic t

Isend

{

non-preemption

Iidle

13/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Burst Limiting Shaper credit evolution

Bursty traffic

t LR LM Transmitted traffic t

Isend

{

non-preemption

Iidle

non preemption

14/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Network calculus

Characteristics of an aggregate traffic of class k crossing the node n

node n Input arrival curve αn

k(t)

Output arrival curve α∗,n

k (t) = αn k(t) ⊘ βn k(t)

minimum service curve βn

k(t)

maximum service curve γn

k(t)

class k

f ⊘ g(t) = sups≥0{f (t + s) − g(s)} f ⊗ g(t) = inf0≤s≤t{f (t − s) + g(s)}

14/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Network calculus

Characteristics of an aggregate traffic of class k crossing the node n

node n Input arrival curve αn

k(t)

Output arrival curve α∗,n

k (t) = αn k(t) ⊘ βn k(t)

minimum service curve βn

k(t)

maximum service curve γn

k(t)

class k t βn

k(t)

αn

k(t)

delay bound

f ⊘ g(t) = sups≥0{f (t + s) − g(s)} f ⊗ g(t) = inf0≤s≤t{f (t − s) + g(s)}

14/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Network calculus

Characteristics of an aggregate traffic of class k crossing the node n

node n Input arrival curve αn

k(t)

Output arrival curve α∗,n

k (t) = αn k(t) ⊘ βn k(t)

minimum service curve βn

k(t)

maximum service curve γn

k(t)

class k t βn

k(t)

αn

k(t)

delay bound βn1

k (t)

βni

k (t)

βnm

k (t)

System S ... ...

βS

k (t) = βn1 k (t) ⊗ ... ⊗ βni k (t) ⊗ .. ⊗ βnm k (t)

f ⊘ g(t) = sups≥0{f (t + s) − g(s)} f ⊗ g(t) = inf0≤s≤t{f (t − s) + g(s)}

15/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Traffic and Network Model

t Traffic modelisation: leaky buckets α(t) The Network Calculus model has been proved in previous worka

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed-Criticality Applications: Model and Timing Analysis”, WFCS 2018

15/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Traffic and Network Model

t Traffic modelisation: leaky buckets α(t) Node modelisation: rate-latency β(t) The Network Calculus model has been proved in previous worka

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed-Criticality Applications: Model and Timing Analysis”, WFCS 2018

15/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Traffic and Network Model

t Traffic modelisation: leaky buckets α(t) Node modelisation: rate-latency β(t)

node sp node bls active BLS BLS class i . . . Non active BLS NBLS class j . . . System mux

The Network Calculus model has been proved in previous worka

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed-Criticality Applications: Model and Timing Analysis”, WFCS 2018

15/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Traffic and Network Model

t Traffic modelisation: leaky buckets α(t) Node modelisation: rate-latency β(t)

node sp node bls active BLS BLS class i . . . Non active BLS NBLS class j . . . System mux βsp

i , βsp j

The Network Calculus model has been proved in previous worka

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed-Criticality Applications: Model and Timing Analysis”, WFCS 2018

15/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Traffic and Network Model

t Traffic modelisation: leaky buckets α(t) Node modelisation: rate-latency β(t)

node sp node bls active BLS BLS class i . . . Non active BLS NBLS class j . . . System mux βsp

i , βsp j

∀ i ∈ BLS, j ∈ NBLS, βmux

i

,βmux

j

The Network Calculus model has been proved in previous worka

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed-Criticality Applications: Model and Timing Analysis”, WFCS 2018

15/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Traffic and Network Model

t Traffic modelisation: leaky buckets α(t) Node modelisation: rate-latency β(t)

node sp node bls active BLS BLS class i . . . Non active BLS NBLS class j . . . System mux βsp

i , βsp j

∀ i ∈ BLS, j ∈ NBLS, βmux

i

,βmux

j

βbls

i ,γbls i

15/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Traffic and Network Model

t Traffic modelisation: leaky buckets α(t) Node modelisation: rate-latency β(t)

node sp node bls active BLS BLS class i . . . Non active BLS NBLS class j . . . System mux βsp

i , βsp j

∀ i ∈ BLS, j ∈ NBLS, βmux

i

,βmux

j

βbls

i ,γbls i

The Network Calculus model has been proved in previous worka

a[WFCS-18] Finzi, A., Mifdaoui et al., ”Incorporating TSN/BLS in AFDX

for Mixed-Criticality Applications: Model and Timing Analysis”, WFCS 2018

16/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Outline

1

System Model

2

Bandwidth Reservation Methods

3

Performance Evaluation

4

Conclusion and perspectives

17/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Problem Overview

Objective Find the reserved BLS bandwidth minimizing RC delay bounds for each flow along its path

17/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Problem Overview

Objective Find the reserved BLS bandwidth minimizing RC delay bounds for each flow along its path Constraints Class rate constraint: in each output port, the class rate is lower than the guaranteed service rate Aggregate rate constraint: the total load of an output port is lower than the total capacity C Deadline constraints: the delay bound of each traffic class is lower than its deadline

18/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Problem Formulation

∀f ∈ RC, ∀mux ∈ pathf , minimize

Lmux

M

,Lmux

R

,BW muxEEDRC,f (Lmux M

, Lmux

R

, BW mux) s.t. ∀f in j ∈ {SCT, RC}, ∀mux ∈ pathf : Rmux

j

• f ∈F mux

j

rf

• g∈F mux

SCT

rg +

• f ∈F mux

RC

rf C Dlf EEDj,f (Lmux

M

, Lmux

R

, BW mux)

18/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Problem Formulation

∀f ∈ RC, ∀mux ∈ pathf , minimize

Lmux

M

,Lmux

R

,BW muxEEDRC,f (Lmux M

, Lmux

R

, BW mux) s.t. ∀f in j ∈ {SCT, RC}, ∀mux ∈ pathf : Rmux

j

• f ∈F mux

j

rf

• g∈F mux

SCT

rg +

• f ∈F mux

RC

rf C Dlf EEDj,f (Lmux

M

, Lmux

R

, BW mux) A complexity of O(lm · N3·m) for m ports and l flows.

19/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Problem Solving

Relaxed Objective Find the reserved BLS bandwidth minimizing RC delay bounds for each class within each output port → Reducing the complexity down to O(m · N3) → Need to define a local Deadline within each output port

19/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Problem Solving

Relaxed Objective Find the reserved BLS bandwidth minimizing RC delay bounds for each class within each output port → Reducing the complexity down to O(m · N3) → Need to define a local Deadline within each output port Solving the problem based on Heuristics The optimisation problem is a non-linear problem Take advantage of conducted sensitivity analysis of the analytical model to deduce heuristics

19/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Problem Solving

Relaxed Objective Find the reserved BLS bandwidth minimizing RC delay bounds for each class within each output port → Reducing the complexity down to O(m · N3) → Need to define a local Deadline within each output port Solving the problem based on Heuristics The optimisation problem is a non-linear problem Take advantage of conducted sensitivity analysis of the analytical model to deduce heuristics Two proposed methods to compute the local deadlines Heuristic Deadline: defined proportionally to the port load Dichotomous Deadline: defined accurately in each port

20/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Outline

1

System Model

2

Bandwidth Reservation Methods

3

Performance Evaluation

4

Conclusion and perspectives

21/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

1-Gigabit Avionics Case study

switch switch switch switch

ES

switch switch switch switch

ES destination ES ES source

Figure: Multi-hop network and traffic communication pattern

21/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

1-Gigabit Avionics Case study

switch switch switch switch

ES

switch switch switch switch

ES destination ES ES source

Figure: Multi-hop network and traffic communication pattern

Priority Traffic type MFS BAG deadline jitter (Bytes) (ms) (ms) (ms) 0/2 SCT 64 2 2 1 RC 320 2 2 3 BE 1024 8 none 0.5

22/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT

URbn

k : bottleneck utilisation rate of class k

22/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT ScenarioSCT =

• URbn

SCT ∈ [0.4 : 80], URbn RC = 20

• URbn

k : bottleneck utilisation rate of class k

22/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT ScenarioSCT =

• URbn

SCT ∈ [0.4 : 80], URbn RC = 20

• 0.5

1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 SCT delay (ms) SCT bottleneck utilisation rate (%)

32 36 SCT deadline

DD Optimised HD Optimised Intuitive 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 RC delay (ms) SCT bottleneck utilisation rate (%)

RC deadline 32 36

DD Optimised HD Optimised Intuitive

URbn

k : bottleneck utilisation rate of class k

22/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT ScenarioSCT =

• URbn

SCT ∈ [0.4 : 80], URbn RC = 20

• 0.5

1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 SCT delay (ms) SCT bottleneck utilisation rate (%)

32 36 SCT deadline

DD Optimised HD Optimised Intuitive 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 RC delay (ms) SCT bottleneck utilisation rate (%)

RC deadline 32 36

DD Optimised HD Optimised Intuitive

→ SCT schedulability is increased by up to 31% under Dichotomous Deadline method

URbn

k : bottleneck utilisation rate of class k

23/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical Results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT

23/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical Results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT ScenarioRC = (URbn

SCT = 20, URbn RC ∈ [0.4 : 80])

23/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical Results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT ScenarioRC = (URbn

SCT = 20, URbn RC ∈ [0.4 : 80])

0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 SCT delay (ms) RC bottleneck utilisation rate (%)

SCT deadline 28

DD Optimised HD Optimised Intuitive 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 RC delay (ms) RC bottleneck utilisation rate (%)

RC deadline 28

DD Optimised HD Optimised Intuitive

23/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical Results

Intuitive parameters: BW = URbn

SCT, LR = MFSRC · BW and

LM = 80 · (1 − BW ) · MFSSCT ScenarioRC = (URbn

SCT = 20, URbn RC ∈ [0.4 : 80])

0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 SCT delay (ms) RC bottleneck utilisation rate (%)

SCT deadline 28

DD Optimised HD Optimised Intuitive 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80 RC delay (ms) RC bottleneck utilisation rate (%)

RC deadline 28

DD Optimised HD Optimised Intuitive

→ RC delay bounds decreased by up to 50% under Dichotomous Deadline method

24/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical Results

improvement compared to SP(%) computation times maximum RC delay at maximum (s) of Scenario URbn

SCT = 33%

URbn

RC = 28%

URbn

SCT

URbn

RC

SCT RC HD BLS 18 22 33 21 57 9 DD BLS 77 55 52 24 117 233

24/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Numerical Results

improvement compared to SP(%) computation times maximum RC delay at maximum (s) of Scenario URbn

SCT = 33%

URbn

RC = 28%

URbn

SCT

URbn

RC

SCT RC HD BLS 18 22 33 21 57 9 DD BLS 77 55 52 24 117 233

→ Higher Complexity for Dichotomous Deadline method

25/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Outline

1

System Model

2

Bandwidth Reservation Methods

3

Performance Evaluation

4

Conclusion and perspectives

26/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Conclusion and prespectives

Two optimized bandwidth reservation methods for TSN/BLS → Heuristic Deadline: simple but average performances → Dichotomous Deadline: complex but good performances Conducted Performance evaluation on a realistic avionics case study → Enhanced SCT schedulability (up to 31%) under DD → Enhanced RC delay bounds (up to to 50%) under DD

26/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Conclusion and prespectives

Two optimized bandwidth reservation methods for TSN/BLS → Heuristic Deadline: simple but average performances → Dichotomous Deadline: complex but good performances Conducted Performance evaluation on a realistic avionics case study → Enhanced SCT schedulability (up to 31%) under DD → Enhanced RC delay bounds (up to to 50%) under DD Approach Generalization to multiple TSN/BLS classes → Offer higher configuration flexibility

27/27 Context and Objectives System Model Bandwidth Reservation Methods Performance Evaluation Conclusion

Q&A

Thank you for your attention